Rodvold K A, Gentry C A, Plank G S, Kraus D M, Nickel E, Gross J R
Department of Pharmacy Practice, College of Pharmacy and Medicine, University of Illinois at Chicago, USA.
Ther Drug Monit. 1995 Jun;17(3):239-46. doi: 10.1097/00007691-199506000-00005.
A dynamic pharmacokinetic model for i.v. vancomycin administration was developed and tested in 47 neonates and infants. Twenty-nine patients (Group 1), having two or more concentrations, were used to estimate population parameters by nonlinear least-squares analysis. Multiple stepwise linear regression techniques showed that estimated creatinine clearance, Clcr, and postnatal age were significant demographic factors related to vancomycin clearance (CL). No strong associations were found for the apparent volume of distribution. A one-compartment model was constructed using the associations of CLcr and postnatal age with vancomycin CL. Eighteen patients (Group 2), receiving 35 courses of vancomycin therapy, with both initial and subsequent sets of peak and trough concentrations, were used to test the predictive performance of the model with and without the use of Bayesian forecasting. Using only population-based parameters, the respective mean error (ME) (bias) and mean absolute error (MAE) (precision) for predicting subsequent peak concentrations were -1.20 and 3.89 mg/L and for trough concentrations, 0.83 and 2.23 mg/L, respectively. For the Bayesian method, these values were, respectively, 0.45 and 4.13 mg/L for peak concentrations and 1.55 and 2.40 mg/L for trough concentrations. When predicted concentrations occurred within 30 days of feedback concentrations, the Bayesian method tended to be slightly less biased and more precise than the population-based parameters. The opposite was true > 30 days of the initial set of feedback concentrations. The use of population-specific pharmacokinetic parameters and Bayesian forecasting should allow accurate dosage regimen design as well as minimize the need for monitoring serum vancomycin concentrations in neonates and young infants.
建立了静脉注射万古霉素的动态药代动力学模型，并在47例新生儿和婴儿中进行了测试。29例患者（第1组）有两个或更多浓度数据，通过非线性最小二乘法分析来估计群体参数。多元逐步线性回归技术表明，估计的肌酐清除率（Clcr）和出生后年龄是与万古霉素清除率（CL）相关的重要人口统计学因素。未发现与分布容积有强关联。利用Clcr和出生后年龄与万古霉素CL的关联构建了单室模型。18例患者（第2组）接受了35个疗程的万古霉素治疗，有初始和后续的峰浓度和谷浓度数据，用于测试该模型在使用和不使用贝叶斯预测情况下的预测性能。仅使用基于群体的参数时，预测后续峰浓度的各自平均误差（ME）（偏差）和平均绝对误差（MAE）（精密度）分别为-1.20和3.89mg/L，预测谷浓度的分别为0.83和2.23mg/L。对于贝叶斯方法，峰浓度的这些值分别为0.45和4.13mg/L，谷浓度的分别为1.55和2.40mg/L。当预测浓度在反馈浓度的30天内出现时，贝叶斯方法往往比基于群体的参数偏差稍小且更精确。在初始反馈浓度的30天以上则相反。使用群体特异性药代动力学参数和贝叶斯预测应能实现准确的给药方案设计，并尽量减少监测新生儿和幼儿血清万古霉素浓度的需求。
